Exa has introduced Agent Ultra, the top effort mode of its Exa Agent API. It is designed for research that needs to explore every possible source. Ultra is used for building large lists, enriching entities, and answering questions with many sources.
Agent Ultra is available as a hosted API. Users can activate it by setting effort to 'ultra' in the API. It is not open-source and cannot be self-hosted. The system splits tasks into smaller parts and assigns subagents to research different domains at the same time.
Ultra uses a mix of frontier models and faster models, choosing the best for each step. It spends the most compute to get the most complete results. Typical tasks finish in about 30 minutes, but very hard tasks can take up to 3 hours.
In tests, Agent Ultra outperformed competitors on four benchmarks. It scored higher on entity finding and data collection, often at a lower cost per task. For example, it found more entities than other systems, with lower costs.
Pricing is based on usage, with a default maximum of $20 per run. Users can set a maximum cost from $1 to $100. Runs that finish early cost less. The API supports streaming and background modes for better control.
This release gives researchers and teams a tool to do more thorough and faster research. It helps improve the quality of large data collections and entity enrichment.
Why it matters
This update offers faster, more complete research results, saving time and increasing data accuracy.
What to do
Try the Agent Ultra API in the Exa API Playground. Set effort to 'ultra' and test it on your research tasks.
